Identifying Accounting Quality
Valeri V. Nikolaev1
The University of Chicago Booth School of Business
[email protected]
Abstract
I develop an econometric strategy that allows identification of accounting quality. The strategy relies on a
new way of characterizing the dynamics of accounting accruals. The characterization is intuitive and does
not hinge on strong assumptions about the earnings and accrual processes, or about managerial
preferences. The identifying assumptions derive from two accounting properties, namely, that both
earnings and cash flows reflect the same underlying performance and that accruals and accounting errors
must reverse over time. My approach discriminates between the accounting error and the part of accruals
that captures the underlying economic performance. The proposed framework also offers a new way of
testing for the presence of earnings management. Implementation issues and empirical evidence are
discussed in a companion paper (Nikolaev 2014).
First draft: June 10, 2014
Current draft: August 8, 2014
1
I thank Phil Berger, Christian Leuz, and Laurence van Lent for providing helpful comments. Financial
support from The University of Chicago Booth School of Business is gratefully acknowledged.
Identifying Accounting Quality
1. Introduction.
In this paper, I develop an econometric framework that allows identification and estimation of
accounting quality. I define accounting quality as the extent to which earnings measure the
(unobservable) underlying economic performance without error. The identification strategy relies on a
new way of modelling the dynamics of accounting accruals, which I am proposing. The proposed
characterization of accruals is intuitive, captures institutional characteristics of accrual accounting, and
does not hinge on strong assumptions about the earnings or accrual processes, managerial preferences, or
other restrictive conditions. The key sources of identification that I exploit derive from our knowledge of
the accounting process and, specifically, from the accounting properties that accruals and accounting
errors must reverse over time, and that earnings and cash flows are both imperfect measures of the same
economic performance.
The quality of accounting information (earnings/accruals) is one of the most widely researched
areas in accounting literature. However, since the key contributions to this area by Jones (1991) and
Dechow and Dichev (2002), the accounting literature has made little progress in advancing the
measurement of accounting quality (e.g., Gerakos 2012). Existing models of earnings quality lack
identification that would allow researchers to differentiate between the “noise” introduced into earnings
by the use of accrual accounting from the volatility of the underlying economic performance (Hribar and
Nichols 2007, Dechow, Ge and Schrand 2010). As a result, the measures of accounting quality based on
proxies for discretionary accruals comingle the variance of earnings explained by accounting error, which
is introduced when measuring economic performance, and the variance explained by the underlying
performance itself, and ultimately the performance volatility or risk.
Another important limitation of existing accruals models is that they attribute all variation in
discretionary (or unexplained) accruals to poor accounting quality and do not take into consideration the
fact that discretionary accruals are also used to facilitate performance measurement. Indeed, it is widely
1
accepted that the primary role of accrual accounting is to improve performance measurement by shifting
the recognition of cash flows over time in a way that reduces cash flow timing problems (AICPA 1973,
Dechow 1994). This implies that a portion of accruals variance is explained by this performance
measurement role and, as such, can be viewed as the “high quality” component of accruals. The existing
literature has not made progress in this regard either and, hence, does not offer a way of identifying the
portion of accruals that facilitates performance measurement. At the same time, identifying such a
component of accruals seems imperative to understanding the economic rationale for accrual accounting.
In this paper, I offer a way to directly address the issues discussed above. I am proposing that
accounting accruals can be described in a parsimonious way by two fundamentally different components,
both of which are distinct from the underlying performance. The purpose of the first component is to
capture the true accrued and deferred portions of the economic performance that are not reflected in cash
flows and thus to facilitate performance measurement (i.e., reduce timing error in cash flows). The second
component of accruals represents “accounting error” or noise introduced into earnings (and accruals)
when measuring the underlying performance. The accounting error can be further decomposed into
intentional and unintentional components (I come back to this issue later).
The simple knowledge that accruals consist of two components does not imply that these
components (and hence characterization of accounting quality) are identifiable. However, it is possible to
exploit the knowledge that, in the longer run, both cash flows and earnings generally agree on the
magnitude of the underlying economic performance, and that both accrual components (performance
component and error component) must reverse over time. This, in turn, motivates a simple dynamic
representation of the accounting accrual process.
The key advantage of this representation of accruals is that, unlike, for example, the model in
Dechow and Dichev (2002), it offers a way of identifying the variance of the two components of accruals
and separating them from the underlying performance, by exploiting information contained in the time
series of accruals, earnings, and cash flows. More specifically, identifying assumptions underlying the
proposed representation, impose structure on a set of moment conditions based on auto- and cross2
correlations in these three time series. Applying General Method of Moments (Hansen 1982), now
available in most statistical packages, to such moment conditions allows for direct estimation of the
variance of the performance measurement components of accruals, i.e., cash flow timing error being
addressed by accrual accounting, the variance of the accounting (or estimation) error in the measurement
of current performance, and the variance and persistence of unobservable economic performance itself.
The key advantage of using GMM estimation over other time series models is that it relies on
minimal distributional assumptions (e.g., it can readily accommodate skewness in earnings).
Nevertheless, before the model can be taken to the data, it is important to make several assumptions about
the correlations among variables in the model. For example, is the accounting error correlated with the
underlying economic performance, ceteris paribus? Can we allow for accounting conservatism to be
present in accruals? Can we allow for earnings management to smooth out shocks to performance?
The framework for modeling accruals and accounting quality offered in this paper is flexible, as it
allows for a number of generalizations. I make a baseline set of assumptions, which are generally
consistent with prior literature, discuss the restrictiveness of these assumptions, and demonstrate how they
can be tested and relaxed in various directions. For example, I discuss how the proposed characterization
can accommodate the asymmetric timeliness of gains and losses (Basu 1997). I also discuss how the
framework can accommodate the modeling income smoothing proposed by Gerakos and Kovrijnykh
(2013) and earnings management more generally.
I contribute to the literature in the following ways. First, I propose an intuitive dynamic
characterization of accruals. Such characterization is compelling due to its simplicity and flexibility,
which does not come at a cost of making restrictive assumptions about the accrual process. To my
knowledge, such representation has not been used in the prior literature. Second, I offer a flexible
identification strategy that can be used to identify the quality of earnings and accruals. To my knowledge,
prior studies do not exploit the idea that cash flows and earnings reflect the same performance to identify
earnings quality. Additionally, prior studies generally do not take advantage of the reversal in accruals to
3
identify accounting quality.2 As a result, models of earnings/accrual quality (e.g., Jones 1991, Dechow
and Dichev 2002) do not offer identification, and comingle the accounting error, the high quality accruals
component, and the underlying economic performance, as I discuss in more detail in Section 5. Third, I
offer an internally consistent framework to identify the variance of performance measurement component
of accruals, i.e., the portion of “high quality” accruals. In contrast to the estimation error, this component
of accruals is central to understanding the quality of accounting information as it reflects the “benefits”
from using accrual basis accounting.
Fourth, I offer several generalizations of the model that allow for the incorporation of various
covariance structures in accrual components. In particular, I offer a way of decomposing the “accounting
error” into unmanaged and managed components and of identifying the variance of these components.
Additionally, I offer a way of modeling the time series of accruals that takes into account the asymmetric
timeliness of gains vs. losses. This is important because Ball and Shivakumar (2006) identify the failure
to accommodate the asymmetry in accruals as an important limitation to the existing models of
discretionary accruals. Fifth, the proposed framework allows researchers to test for the presence of
earnings management and overcomes limitations of alternative approaches that rely on proxies for
discretionary accruals (Gerakos 2012).
Finally, the study is related to several recent papers that offer ways to overcome some limitations
of standard discretionary accruals models. Gerakos and Kovrijnykh (2012) advocate indirect inference
about earnings management by testing for negative second-order auto-correlation in the error term in an
earnings auto-regression. Beyer, Guttman and Marinovic (2014) rely on a structural model that makes
assumptions about managerial preferences as a function of stock-based compensation to characterize
dynamic equity valuation. The model does not account for the reversal in accruals but instead relies on
2
Two prior studies that exploit reversal are Dechow et al (2012), and Gerakos and Kovrijnykh (2013).
Dechow et al use reversal primarily as a way to gain statistical power to detect earnings management rather than as
an identification tool (identification comes from the specification of exogenous partitioning variable), although they
also point out that reversal improves model specification. Gerakos and Kovrijnykh focus on the reversal of earnings
management and do not model accruals or earnings quality more generally.
4
structural equations and the effect of earnings management on stock returns to identify accounting
quality.
Note that, due to a potentially large number of ways to relax the baseline set of assumptions, I
leave the discussion of implementation issues and empirical evidence for a companion paper (Nikolaev
2014). The current study proceeds as follows. The next section develops a dynamic representation of
accounting accruals and discusses accounting quality in this context. Section 3 discusses the identification
strategy and lays out a baseline set of assumptions. Section 4 considers ways to relax these assumptions.
Section 5 more formally discusses the advantages of the current approach to identifying accounting
quality over conventional approaches. Section 6 considers the effect of accounting conservatism. Section
7 discusses explicit ways to account for earnings management. Section 8 concludes the study.
2. A New Representation of Accruals.
In this section, I propose a simple dynamic representation of accounting accruals. This
representation is based on the idea that cash flows and earnings can be viewed as noisy measures of the
underlying economic performance (e.g., Stickney and Weil 2007). This view, in turn, implies a simple
dynamic structure for accounting accruals, which can be exploited to understand and identify earnings
(accruals) quality. Consistent with the Statement of Financial Accounting Concepts No. 1 and Dechow,
Ge and Schrand (2010), I define the quality of accounting earnings (accruals) in the following way:
Definition: Accounting quality is the extent to which accrual accounting facilitates the
measurement of the underlying economic performance.
In the rest of this section, I develop this concept more formally.
2.1. Accounting error in periodic performance measures.
How well did a company perform over a period of time? There is no perfect way to answer this
question. Indeed, the need to measure economic performance over relatively short time periods, such as
quarters or years, is ultimately at the root of measurement problems in accounting. Different performance
measures, such as accounting earnings, cash flows, or stock returns, will generally give the same or
5
similar answer about a firm’s performance in the longer run (e.g., Trueblood study, AICPA 1977, Easton,
Harris and Ohlson 1992).3 In the short run, however, these measures will exhibit a different degree of
measurement error. The larger the measurement error, the lower the quality of an accounting performance
measure.
The measurement error across different performance measures has a key similarity: it arises
because of the difficulties in attributing (allocating) the underlying economic performance to a particular
time period. Thus, it can be thought of as a “timing error” that cancels out over time and thus exhibits
negative autocorrelation. To make things more concrete, suppose
is the true underlying economic
performance over a period . One way to measure this unobservable economic performance is to rely on
cash flow from operations,
. However, such performance measure will contain noise due to timing and
matching problems present in cash flows (Dechow 1994). These timing and matching problems result in
measurement error
that reverses in the future. This idea is modeled econometrically using the
following representation of cash flows:
,
(1)
which captures the idea that cash flows are a noisy measure of a firm’s performance. As the performance
measurement error reverses over time, long run cash flow will approximate the long-run economic
performance
. To the extent that accounting earnings are designed to improve performance
measurement over that implied by cash flows, the structure of the error term in equation (1) yields a
natural role for accounting accruals. Specifically, to measure the underlying performance properly,
accountants need to book an accrual
to offset the timing error in cash flows.
However, even if accountants observe the true performance of the firm and know
, in order to
account for this component they will generally have to follow a certain set of procedures that introduce
assumptions and require the use of estimates (often based on prior history). As a result, accruals will
3
For example, a simple aggregation of cash inflows less cash outflows over a number of periods will also do a good
job in reflecting the underlying economic performance. At one extreme, the total net cash inflow equals the true
economic performance over the life of the firm. 6
introduce an accounting error into earnings, which will, presumably, be smaller than the timing error in
cash flows. Nevertheless, the error will share the same econometric properties and, hence, the accounting
earnings will have a similar econometric structure:
.
(2)
As in the case of cash flows, the measurement error
also exhibits reversal over time. However,
the reversal arises due to a different reason, namely because accrual accounting relies on estimates and
assumptions to measure the underlying performance. Over longer horizons, earnings will also measure the
economic performance with a high degree of accuracy because the timing errors cancel out over time.
2.2. The implied characterization of accruals.
The econometric representation of earnings and cash flows implies a simple (and intuitive)
dynamic characterization of accounting accruals. Since accounting accruals
are equal earnings less
cash flows, we have:
.
(3)
≡
I refer to the first component of accruals,
, as a “performance measurement
component” of accruals because the role of this component is to facilitate the measurement of underlying
economic performance relative to cash flows. In other words, because
as “high quality” portion of accruals (in the sense that
performance not captured by cash flows. Note that
, we can also think of
, i.e., a portion of economic
should not be viewed as non-discretionary
accruals, as a firm’s management may have the full discretion to choose the level of accruals that capture
the underlying performance correctly but will generally do so with error.
The second component of accruals,
≡
, arises because of imperfections in accrual
accounting. I refer to this component as the “accounting error” or the “estimation error,” which is in line
with the terminology in Dechow and Dichev (2002). Empirically, I expect the variance of
average, to be smaller than the variance of
, as otherwise the rationale for accrual accounting is
unclear.
7
, on
The above representation of accruals captures the salient institutional properties of accrual
accounting. First, using conventional accounting terminology, the component
represents deferred
performance (deferred revenue less deferred expenses) less accrued performance (accrued revenue less
accrued expenses) in the absence of accounting noise. This is intuitive, as deferred revenue (expenses)
implies higher (lower) cash flow relative to underlying performance, whereas accrued revenue (expenses)
implies lower (higher) cash flows relative to the underlying performance. In turn, the component
represents the reversals of accruals or deferrals previously created.4 Second, accruals exhibit reversal over
time. Over longer horizons, both cash flow and earnings are equivalent in measuring economic
performance. Third, the role of accruals is to mitigate the known problems in cash flows and hence to
facilitate performance measurement (Dechow 1994). Fourth, any gain in performance measurement from
accruals comes at the cost of using estimates and assumptions, which introduce an estimation error in
accruals and earnings (Dechow and Dichev 2002).
2.3. Understanding accounting quality.
The above representation reveals that the variances of cash flows, earnings, and accruals are
jointly determined by the following parameters:
first component
,
, and
. The
is the volatility of the underlying economic performance. This component can be
thought of as the riskiness of the business in which a company operates or as operating volatility (a
portion of this risk can be idiosyncratic and the rest can be systematic, i.e., exhibit links to market beta,
etc.).
The parameters
and
are the variance of performance component of accruals (variance of
high quality accruals), and the variance of the estimation error, respectively. These components, jointly,
determine accounting quality. In particular, the component
as discussed previously, whereas the component
captures the benefit of accrual accounting,
reflects the cost of accrual accounting. Note that the
4
For example, if a customer makes a $1 advance payment in the current period (for work performed in the future),
operating cash flow overstates performance by $1 in the current period and will understate performance by the same
amount in the future. Subsequently, a negative $1 accrual is booked upon a receipt of a $1 advance payment. 8
former is equally important as the latter in understanding “earnings quality” as well as the demand for
accrual accounting more generally.
The above discussion reveals a gap in the literature. Specifically, prior research, following
Dechow and Dichev (2002), focused on quantifying and understanding the variance of
component and not the variance of
. Indeed, the literature has so far neglected the benefits of
accrual accounting, i.e., the variance of the
component of accruals. More importantly, the above
suggests that focusing on the variance of the estimation error results in an incomplete picture of
accounting quality, as it does not take into account the benefit side of accruals. Consistent with the
definition of earnings quality, one way of constructing an overall measure of accounting quality would be
as follows:
AccountingQuality ≡
.
The numerator of this ratio captures the net benefit of using accruals, which is scaled by the magnitude of
the error introduced by accrual accounting. Intuitively, earnings are of high quality when the variance of
the performance portion of accruals (i.e., high quality accruals) is high relative to the variance of the
estimation error and vice versa. Depending on the research objective, one may also be interested in the
following two signal-to-noise ratios:
r
/
E ≡
, and r
/
C ≡
.
These ratios reflect the intuition that the performance of riskier companies is harder to measure.
3. Identification of Accounting Quality.
My identification strategy relies on the time series properties of the above model. Intuitively, the
primary source of identification comes from the fundamental properties of the accounting process, and
namely, from our knowledge that (1) accounting accruals and estimation error reverse over time, and (2)
that both earnings and cash flows are noisy measures of the same underlying performance. To illustrate
9
this, first, consider a hypothetical example where earnings process is
and
,
, and
are
independently distributed random variables. In such a case, it would not be possible to identify the
variances of
,
or
unless one used additional information to separate the measurement error from
the performance. Such information is available in the structure of the error term and in cash flows. Indeed,
, the reversal can be used as an identifying assumption based on
if the error process is
which the variance of
and
components can be recovered. Specifically, the resulting moment
2
conditions E
, and E
represent two equations with two unknowns that
can be solved for using GMM (or other minimum distance methods). Similarly, our knowledge that
offers a moment condition, E
, that, combined with E
, can be used to
identify the variance of estimation error, etc. Note that, while in this simple example either of the
identifying assumptions will suffice, both will contribute towards identifying the parameters in more
general cases.
Following the intuition above, I choose moment conditions that exploit reversal in both the
performance component of accruals (high quality accruals) and the estimation error, and also that
combine the information in earnings and cash flows. Constructing moment conditions from multiple time
series comes in handy when time series are relatively short, as it exploits more information.
3.1. Identifying assumptions.
The model (1) does not rely on any restrictive assumptions about the accrual process. The
identifying assumption made so far is that the reversal of accruals happens in the next reporting period. In
this section, I state more formally a flexible set of assumptions that allow taking the model to the data.
While these assumptions are consistent with prior literature, they should be viewed as a benchmark case.
Assumption 1. (i)
is a stationary process; (ii) timing errors
and
reverse within subsequent
periods.
,
Assumption 2. (i) cov
Assumption 3. cov
,
0, and (ii) cov
0 for all .
10
,
0 for all
1.
,
Assumption 4. (i) cov
0, and (ii) cov
,
0 or all .
This baseline set of assumptions warrants some discussion. The assumption (1) is generally
descriptive of working capital accruals on an annual basis (Dechow and Dichev 2002). These accruals are
of primary interest to the literature on earnings quality. The assumption (2) is warranted because the error
structure (e.g.,
) directly incorporates the expected negative autocorrelation in accruals.
Assumption (3) is also plausible institutionally. Recall that accountants measure the true accrued/deferred
portion of economic performance
expectation error
to offset the timing error in cash flows and in doing so introduce an
. If accruals measure performance in an unbiased way, there should not be a
systematic relation between the error
being corrected and the error
that arises in doing so.
Assumption (4) also seems plausible. For example, Dechow and Dichev (2002) make an implicit
assumption that the estimation error and performance are uncorrelated: cov
,
0. Further, it is not
clear why the timing error in cash flows should be expected to systematically correlate with performance.
For example, in a steady-state firm one would not expect that a positive shock to economic performance
is more likely to result in the need for accrued performance (negative
performance (positive
, as opposed to deferred
).
The main advantage of using the assumptions (1)-(4) is the parsimony of the model. It is
important to understand how these assumptions can be relaxed because a violation of these assumptions
will be generally informative of accounting quality. In subsequent sections of the paper, I will consider
generalizations of the model and offer ways to relax the baseline assumptions in a number of ways.
3.2. Moment conditions.
Recall that the processes for earnings, accruals, and cash flows have the following dynamic
representation (for convenience stated in one place):
(4a)
(4b)
.
(4c)
11
This representation, in conjunction with assumptions (1)-(4) stated above, allows for identification of the
variance of the performance component of accruals,
and the variance of shocks to economic performance,
, the variance of accounting estimation error,
,
. Doing so requires selecting moment conditions
that are informative about the parameters of interest. Note that the choice of moments is not guided by
econometric theory and is more of an art than science. The natural principles that should guide the choice
of moment conditions: (1) moments should provide substantial information about the parameters of
interest, i.e., changing a parameter should have a first-order effect on the moment; (2) moments that rely
on assumptions that are unlikely to be met should be avoided, and (3) moments that are linearly
dependent should not be used (e.g., note that moment E
E
and E
does not add information if moments
are used). I suggest the following simple set of moments conditions, however, note
that other moments can be used as well (for example, using the moment E
assumptions about the process
). Note that all variables are measured as deviations from their means.
m 1 : E
2
,
m(2 : E
2
,
m(3):
E
2
2
,
m 4 : E
,
m(5 : E
,
E
,
m(6):
where
requires additional
is the first-order autocorrelation of economic performance
, i.e., persistence of the underlying
economic performance.
These simple moment conditions present six equations with four unknowns (i.e., the model is
overidentified). The first three equations help identification by exploiting the information in two different
performance measures, whereas the last three moments also exploit the reversal in the error components.
GMM estimation can be used to identify and consistently estimate the parameters
12
,
,
, and
at
firm or industry level. Maximum likelihood can also be sued to estimate these parameters but requires
strong distributional assumptions, such as normality, which is an important advantage of using GMM.
As an alternative, we can use changes in the time series of earnings, cash flows, and accruals. In
this case, it is important to account for differences in the dynamics of the three time series. For example,
the differenced version of the accrual process given by equation (4b) is as follows:
2
2
.
Differenced formulation implies the following set of moment conditions:
m 1 : E
∆
6
,
m(2 : E
∆
6
,
E
6
m(3):
6
,
m 4 : E
∆
∆
4
,
m(5 : E
∆
∆
4
,
4
,
m(6):
E
where
∆
4
is the first-order autocorrelation of changes in the economic performance,
.
Such formulation has two principal advantages over the levels formulation above. It no longer requires
that the process
is stationary (only its changes must be stationary), i.e., this formulation allows for a
very weak assumptions about economic performance. Second, differencing effectively eliminates highly
persistent, i.e., long-term accrual components, such as depreciation or amortization (e.g., of bond
premium).
The proposed empirical strategy has a principle advantage over the models based on Jones (1991)
or Dechow and Dichev (2002). Namely, these models do not allow for identifying and separating the
variance of estimation error and the variance of high quality accruals while controlling for operating
volatility. I discuss this in more detail in Section 5, where I also offer reconciliation between the
representation of accruals offered here and that in Dechow and Dichev (2002).
13
4. Model Generalizations.
In this section, I discuss how assumptions (1) and (4) can be relaxed to incorporate potentially
more realistic scenarios. Relaxing these assumptions should be decided on a case-by-case basis as it
presents a trade-off: a potentially better fit of the model with the data comes at a risk of overparameterizing the model (without a significant gain in explanatory power). Over-parameterizing the
model is of particular concern when the number of observations is relatively small, e.g., when the
estimation is done at a firm level.
Several directions to relax the baseline assumptions require particular consideration. First, the
timing errors
and
may be auto-correlated. For example, a positive auto-correlation in
may arise
because cash flow tends to systematically fall short of economic performance in the growth phase. The
estimation error
may be auto-correlated because consistent application of accounting procedures and
assumptions is likely to result in a correlated error in performance measurement. Second, the components
and
may be cross-correlated if accounting error tends to be correlated with the timing error. This
may take place due to the presence of accounting conservatism to the extent that unrealized profits are
conservatively deferred until the future. Finally, the shocks to the underlying economic performance may
be correlated with accounting error (Gerakos and Kovrijnykh 2013) or cash flow timing error. To allow
for these possibilities, I consider an alternative set of assumptions, discussed next.
4.1. An alternative set of assumptions.
Relaxing assumptions (1) and (4) generally requires imposing more structure on the process
(so far we have only assumed stationary). For example, modeling non-zero covariance between the
accrual components and lagged values of economic performance makes it necessary to specify process
One way to do this issue is to modify assumption (1) as follows:
Assumption 1A. (i)
1,0 process,
follows
performance; (ii) timing errors
and
reverse within the subsequent period.
14
, where
is a shock to
.
This assumption is not particularly important here but is also made in Gerakos and Kovrijnykh
(2013) and Beyer, Guttman and Marinovic (2014). I use the following convenient notation (recall that
under the initial set of assumptions,
≡ cov
,
≡ cov
≡ cov
,
,
and
are uncorrelated with
,
≡ cov
, and
≡ cov
,
,
):
,
,
.
The alternative set of assumptions that accounts for issues discussed above is as follows :
Assumption 2A. (i)
, and
Assumption 3A.
Assumption 4A. (i)
for
0 for
, and
, and
1; (ii)
0 for
0.
0 for
0; (ii)
1.
0
, and
0.
Under this set of assumptions, we have nine parameters of interest (
,
0 for
, and
,
,
, ,
,
,
,
) and thus we require at least nine moment conditions that are linearly independent. For
,E
example, one could use an analogous set of equations E
, and E
, with
taking values from 0 to 3, which amounts to 12 moments. Ideally, we would also want to exploit the time
series of
but this is not feasible since the underlying performance cannot be observed directly.
4.2. Using information in the time series process of economic performance.
If performance
was observable, it would be possible to use the time series of
to construct
additional moment conditions and gain additional information. For example, using moments E
running an OLS regression
and
) or
would be a powerful way of identifying the parameters
. While this is not feasible, it is possible to take advantage of this idea by substituting
into equation
to express earnings as a time series process with autoregressive
components (an analogous substitution is used in Gerakos and Kovrijnykh 2013). Similarly, we can
into the same equation to obtain the following set of equations that no
substitute
longer depend on
.
1
,
(5a)
15
1
,
(5b)
.
(5c)
There are a number of ways how these equations can be used to formulate moments. For
or
example, if we multiply both sides of equations (5a) and (5b) by
and take expectations we
will obtain a set of moments representing analogues of Yule-Walker equations. Note, further that
0, which
equation (5c) is no longer a linear combination of the other two equations for parameters
generally increases the number of moments. One convenient way to formulate moment conditions is as
follows:
Auto-covariance of earnings.
m(1 : E
)(
m(2 : E
)(
m(3 : E
)(
2
2
1
2
1
1
1
,
2 1
1
,
.
Auto-covariance of cash flows.
m(4 : E
)(
m(5 : E
)(
m(6 : E
)(
2
2
1
2
1
1
1
,
2 1
1
,
.
Cross-covariance of earnings and cash flows.
m(7 : E
)(
m(8 : E
)(
m(9 : E
)(
2
1
1
)(
m(11 : E
)(
1
1
.
Auto-covariance of accruals.
m(13 : E
2
2
4
2
2
2
2
,
2
16
,
,
,
m(10 : E
m(12 : E
1
,
,
m(14 : E
.
It is important to note that, while assumption about the process
adds statistical power and
complements identification strategy, it is also a potential source of misspecification. This assumption can
be relaxed by, for example, including more lags into the autoregressive process or adding a moving
average component. This, however, comes at a cost of adding more parameters and algebraic complexity.
In such a case, the moment conditions should be altered accordingly. Another way to relax assumption
(1A) is to explicitly model non-stationary
(note that non-stationarity is less likely to be an issue for
earnings scaled by total assets; the analogy would be stock returns). A conventional way to do this is to
as ARMA process. In such case, the moment conditions considered
model the first difference
above need to be modified in order to accommodate for the differenced time series (see p. 13).
4.3. Short-term vs. long-term accruals.
In accordance with Assumption 1, the focus so far has been on working capital accruals, which
are expected to reverse within one year. It is also useful to consider a possibility of a different reversal
patterns. For example, quarterly accruals will generally exhibit reversals over several quarters. Nonworking capital accruals, such as capital expenditures (capital expenditures are not a part of operating
cash flows also reverse over a number of periods in the form of depreciation or future write-offs.
One approach to deal with long-term accruals in the literature is to try to exclude them from
earnings (e.g., use earnings before depreciation). Nevertheless, my proposed framework can be used to
model more complex reversal patterns. To make things more concrete, suppose a portion of accruals
reverses in period 1 and the remainder reverses in period 2. In that case, we have the following structure
of the model:
1
,
1
(6a)
1
1
,
.
(6c)
17
(6b)
This model adds an additional parameter
that governs the reversal process and can be estimated. These
equations can be used to construct moment conditions. For example, under the baseline set of
assumptions the following moments will hold:
m(1 : E
2
1
,
m(2 : E
1
m(3 : E
.
,
The model can be generalized further by allowing
, reverse over
with certain probability. For example, one reversal pattern to consider is
periods at a certain rate or
1/
…
1/
,
which is applicable to non-current accruals (e.g., improper capitalization of PPE and subsequent
depreciation). Notice that the distribution of
(and other variables in the model) can be asymmetric (e.g.,
as in the case of capital expenditures) and the estimation method does not make distributional
assumptions. A slightly different approach to dealing with long term accruals would be to allow for
accrual components that reverse in
. I view these extensions of the baseline
periods, i.e.,
model as opportunities for future research.
5. Distinctions from Prior Approaches.
In this section, I discuss the main distinctions of the approach to modeling accounting quality
proposed here from more conventional approaches in the literature. I discuss why the conventional
approaches do not identify the variance of the estimation error or accounting quality.
The two primary approaches to modeling accruals and estimating earnings/accruals quality
available in the literature are based on important contributions by Jones (1991) and Dechow and Dichev
(2002). Both the Jones model and Dechow-Dichev model use a linear regression model to separate
accruals into their expected and unexpected components, often loosely referred to as non-discretionary
and discretionary accruals, respectively. The explained variable in both models is accruals and the
explanatory variables include change in revenue and PPE (Jones model), and current, future, and past
18
cash flow from operations (Dechow-Dichev model). The standard deviation of residuals from these
models (or their absolute values) are used in the literature to proxy for accruals/earnings quality (e.g.,
Francis et al., 2004). The primary concern with these models is that they do not properly control the
(unmanipulated) economic performance when determining the “discretionary” portion (e.g., Dechow et
al., 2010) and as a result are picking up the volatility of the working capital and ultimately the riskiness of
the firm. The approach here directly overcomes this issue by explicitly separating risk and the
performance component of accruals from estimation error. To be more concrete, I discuss this in more
detail in the context of the Dechow-Dichev model.
The main objective of the Dechow-Dichev model is to estimate the variance of the estimation
error (which may or may not be subject to manipulations). The model is based on the following set of
equations (Dechow and Dichev 2002, pp. 38-40), using my notation:
,
(7a)
,
(7b)
.
(7c)
is cash received in period and recognized in earnings in period , and other variables are
where
defined previously. Intuitively, the model states that earnings are the sum of current, past, and future cash
flows recognized in current period earnings, plus an estimation error
and its reversal
. In contrast,
cash flows represent a portion of past, current, and future earnings realized in the current period. Accruals
are thus the difference between the earnings and cash flows.
Note that such representation of accruals appears quite different from the representation
advocated in this paper, which raises the question of which representation is more realistic. However, it is
straightforward to show that the two representations reconcile.5 Indeed, one can add and subtract terms
and
to equation (7a) so that both earnings and cash flows include the portion
.
This portion represents ultimate cash implications of events that occur in the current period and hence is
5
A disadvantage of representation offered in Dechow and Dichev (2002) is that the variance of component
not appear to be feasible to identify, unlike the variance of .
19
does
. Further, we can define the
equal to true performance in the current period, i.e.,
following variable
≡
and its lagged counterpart
≡
. It is
straightforward to show that substituting these definitions into equations (7a), (7b), and (7c) and
rearranging them obtains the set of equations (4a) (4b) and (4c). The accruals component
an analogous interpretation in this context. The portion
has
is a portion of current earnings realized
(collected/paid) in the future less a portion of current cash flow recognized (earned/incurred) in the next
, is the cash flow collected in the prior period but recognized
period. The second component,
currently less cash recognized previously but collected in the current period.
, characterized by parameter
To estimate the variance of the estimation error, i.e.,
, Dechow and Dichev (2002) use the variance of the residual from the following regression model
(constant is omitted) motivated by equation (7b):
.
(8)
It not difficult to see that the variance of the residual
error and, in fact, comingles
falls short of identifying the estimation
with the variance of high quality accruals,
the underlying economic performance,
, as well as the variance of
. To see this more explicitly, it is convenient to express
equation (8) in terms of the underlying performance
by substituting equations (1) and (2) into equation
(8):
.
(9)
Examination of the dependent variable reveals that to isolate the variance of the estimation error
, it is necessary to control for the performance component of accruals
(in which case
). For this to happen, it must be the case that:
.
This is a contradiction as the equality (9) does not hold under any set of parameters in the linear model.
As a result, the residual,
, will contain an additional term,
20
, so that:
,
.
Because the variance of
is a function of
and
, the regression (9) does not isolate high
quality accruals: the variance of the residual thus becomes a function of risk,
of accruals,
, and the variance of the estimation error
, the performance portion
. This mimics the conclusion in Wysocki
(2009) in that the current regression model does not separate “high quality” and “low quality” accruals.
A similar logic applies to the Jones model as, by controlling for change in revenue and the level
of PPE, this model will generally not fully control for the performance component of accruals. In sum, the
existing models do not identify the variance of the estimation error in earnings/accruals. Furthermore,
these models are not suitable to identify the variance of the performance measurement component of
accruals.
6. Incorporating Accounting Conservatism.
Ball and Shivakumar (2006) argue that, in addition to mitigating the transitory variation in cash
flows (Dechow 1994), the (performance measurement) role of accruals also includes the accelerated
recognition of unrealized economic losses and gains that arise due to changes in expectations of future
cash flows. Note that unrealized gains and losses can also be viewed as timing errors in cash flows and
hence fall under
. The key difference, however, is that the unrealized losses and gains follow
asymmetric recognition under the conservative accounting rules (Basu 1997). Failure to take this into
account constitutes an important criticism of the existing models of discretionary accruals (Ball and
Shivakumar 2006) which, for example, are likely to classify write-offs and impairment charges as
“discretionary accruals.” Such a classification is problematic because these charges facilitate the
recognition of adverse economic performance and in this sense are a desirable property of financial
reporting. It is necessary to point out, however, that asymmetrically recorded gains and losses are less
21
likely to be a part of working capital accruals (because write-offs and impairments usually primarily
affect non-current accruals, e.g., PPE, and are reported separately).
The way of identifying accounting quality I am proposing can accommodate asymmetric
timeliness of gain and loss recognition. It also allows identifying and isolating a portion of accruals
explained by conservatism. Let me first illustrate this with a simple example. In particular, let’s hold other
things equal and consider a loss
$100 that happens in the current period but is not realized until a
$100 and requires a negative
future period. Such a loss implies the timing error in cash flows
accrual of
$100 to offset the timing error. Note that accruals improve performance measurement
and are not classified as “discretionary,” i.e., a part of accounting noise. By analogy, an unrealized gain
$100 implies an equivalent timing error in cash flows
$100 (again because the gain is not
$100 should be
realized until the future). Without a conservative treatment of gains, an accrual
recorded. Because a conservative treatment of gains does not consider $100 as current income, it
introduces a conservative accounting error
$100 into earnings (and accruals). Intuitively,
accounting conservatism introduces skewness into accruals and earnings as it implies that accountants are
more likely to make negative errors than positive ones when measuring performance (and hence less
likely to report high positive earnings).
The above example has two main implications. First, economic performance (through unrealized
gains/losses) is now correlated with the timing error
. Second, and more importantly, the asymmetric
treatment of gains and losses introduces a nonlinear relation between
and
. To allow for such non-
linearity in the accrual process we can generalize the model in the following way:
,
,
g
,
,
22
where
and
represent the timing error in a sense discussed previously (Dechow 1994), and
exclude unrealized gains and losses that are treated asymmetrically. The component
represents
unrealized gains (losses) that happen in the current period and thus is subject to asymmetric treatment,
but falls out from current cash flow due to its timing error
which is a part economic performance
Because of asymmetric treatment of gains and losses, g
.
is expected to be close to 0 for negative news
(because negative news is recognized immediately). In contrast, g
is expected to be equal to
, as
long as unrealized gains are deferred until realization (if gains are deferred only partially, we would
expect g
, where 0
1).
The natural candidate for the function g(.) is g
max 0,
(the case when we have unrealized loss), the accrual
the estimation error, max 0,
. Intuitively, when
is recorded and the corresponding part of
, is zero. In contrast, when
is positive (the case of unrealized gain),
, which implies
conservative accounting precludes recognition of
is negative
0. To
and
construct moment conditions, it necessary to add assumptions about the distribution of
. To keep the set
of moment conditions simple, I keep assumptions (1)-(4) and make the following assumption:
Assumption 5: The distribution of
and
is symmetric around 0,
,
0,
,
0
0 for all s.
,
In such case, the following conditions become relevant:
E
,
E max 0, z
E
E
E
0.5
max 0, z
E
E
z
E
E
max 0, z
0.5
,
,
,
0.5
.
These modifications introduce one additional parameter,
, and can be easily incorporated into
the moment conditions discussed above. These moment conditions allow for estimating and isolating the
23
portion of accruals explained by accounting conservatism. We are still able to identify the variance of the
remaining components of accruals.
Note that the symmetry assumption can be dropped at the expense of one extra parameter,
. In
such a case the following modifications take place:
E max 0, z
E
E
E
max 0, z
E
E
,
,
max 0, z
.
I view these extensions of the model as opportunities for future research.
7. Identifying Earnings Management.
So far I have not made assumptions about whether the estimation error is an unintentional error or
whether it has a “managed” component. In this section, I discuss how the framework offered in this paper
can be used to accommodate earnings management. I also discuss how one can use the model to test
hypotheses about the presence of earnings management, while fully taking into account the dynamics
accrual components.
To allow for earnings management, it is possible to extend the model to include the following
structure of accounting error:
,
where
(10)
is the unmanaged portion of the estimation error and
is its managed component. To separate
the managed component, however, it is necessary to either (1) impose assumptions on the dynamic
process through which earnings management is manifested in accruals (e.g., Gerakos and Kovrijnykh
2013), or (2) rely on additional information about the incentives to manage earnings (e.g., Dechow et al.
1995). I discuss these alternatives in the remainder of this section.
7.1. Income smoothing.
Gerakos and Kovrijnykh (2013) argue that earnings management is manifested via performance
smoothing. They show that incentives to smooth out shocks to economic performance arise under a fairly
24
general set of assumptions about managerial preferences (e.g., meet-or-beat targets). In line with this
intuition, they model earnings using the following dynamic structure:
,
(11)
where true (unmanaged) economic performance,
, is assumed to follow an AR(1) process,
. Based on this assumption, the authors predict that earnings will also follow an autoregressive
process, however, with a negative second-order autocorrelation of the error term. Their estimate of the
second-order autocorrelation parameter confirms this prediction.6
Note that the above earnings representation relies on a rather strong assumption that, effectively,
0 and
, i.e., the entire estimation error in accruals is explained by earnings
management. The current framework allows for relaxing this assumption while at the same time
incorporating the idea in Gerakos and Kovrijnykh (2013) that earnings management introduces a
correlation between economic performance and accounting error. This can be achieved by allowing
estimation error to incorporate both an unmanaged error as well as the income smoothing component:
.
Under the assumptions (1)-(4), it is necessary to revise the moment conditions to incorporate the
following covariance structure:
E
E
,
.
As previously, such an approach allows for identifying the variance of the estimation error as well
as other parameters of interest. The advantage of using this approach in conjunction with GMM
estimation is that the parameter
can be estimated directly (as opposed to being imputed from the
second-order autocorrelation in the error term of the reported earnings), and without imposing a normality
assumption necessary to estimate ARMA process. In addition, GMM takes into account the information
about
contained in the first-order autocorrelation of the error term in reported earnings.
6
They further impute the parameter , which indicates significant earnings management (on average, the
parameter is 0.32, which implies that roughly 1/3 of performance shocks (and hence ultimately performance) is
being “hidden” via earnings smoothing). 25
7.2. Directional tests of earnings management.
Another way to isolate earnings management in accruals is to model it as a function of the
underlying economic incentives to manage earnings. Suppose we have a variable
, given exogenously,
which is informative about managerial incentives to overstate (understate) earnings and accruals (i.e.,
estimation error). For example,
could be proximity to covenant violation or some other threshold or
benchmark (e.g., earnings forecast). Alternatively,
could be a dummy variable that takes the value of
one in firm-years where the incentives for manipulation are present (e.g., Dechow, Sloan and Sweeney
1995). In this case, we can model that the estimation error has the following process:
,
(12)
where the first component captures the extent of earnings manipulation as a function of the economic
, and the second component captures the portion of the estimation error
incentives, i.e.,
orthogonal to incentives to manage earnings.
To start, I will assume that
is independently distributed over time and is also independent of
other variables in the model. To avoid unnecessary parameters, let’s normalize the mean of
to zero and
1.
Under the baseline set of assumptions, the following modifications to the moment conditions
stated earlier are necessary to take into account:
E
E
,
.
Note that these two equations with two unknowns can be solved as a part of a larger set of moments to
decompose accruals into both managed and unmanaged portions. Furthermore, it is possible to test
whether
and, hence, earnings management, is statistically significant and in the predicted direction. This
could be done by pooling cross-sectional and time series variation (to the extent that a cross-sectional
pool has similar firms, e.g., at industry level, or to the extent that we allow for heterogeneity in other
examples of the model parameters). In this sense, this approach complements the approach in Dechow,
26
Sloan and Sweeney (1995) and Dechow, Hutton, Kim and Sloan (2012), who model discretionary
accruals as a function of
to test for earnings management. In particular, testing for
0 explicitly
accounts for reversal of earnings management on a different period, which is something that Dechow et
al. (2012) have shown to increase statistical power.
However, the current approach has several advantages. Gerakos (2012) argues that discretionary
accruals based on, e.g., Jones (1991) do not properly account for the dynamic properties of unmanaged
accruals. In particular, my framework reveals the following two issues. The first issue arises if incentives
to manage earnings and the timing error in cash flows are correlated, which can be expressed as follows:
,
(13)
Indeed, there are reasons to expect that the incentives to manage earnings can be correlated with a
set of factors, e.g., growth, that in turn are correlated with
accruals do not properly isolate the
. Because the proxies for discretionary
component, their use will generally result in confounded
inferences with respect to earnings management. To simplify things, suppose one uses total accruals to
test for earnings management and runs the following regression:
.
The estimated slope coefficient in this regression has the following expected value:
,
E
.
Thus, depending on the sign of the coefficient of
, a regression framework can overstate or
understate the estimated magnitude of earnings management,
moments to estimate and isolate
by making use of following two moments:
, and E
E
. In contrast, we can use the method of
.
The second problem arises if the incentives to manage earnings persist for several periods, e.g.,
0, now we have:
. Assuming for simplicity that
E
E
.
27
In other words, the more persistent earnings management is, the harder it is to detect it within a
regression framework, i.e., without accounting for persistence in
. Intuitively, this takes place because it
is difficult to consistently overstate accruals for a number of periods in a row. In contrast, the following
moment conditions make use of lags of
management
to overcoming this problem and identify the degree of earnings
:
E
, and E
.
To sum up, the framework I am proposing, in contrast to those that rely on proxies for
discretionary accruals, is a one-step estimation immune to the issues discussed above because it directly
accounts for the dynamics of the managed and unmanaged accruals process.
8. Conclusions.
The quality of accounting earnings and accruals is one of the most actively researched areas in
accounting. Over the past decades, however, the literature has made rather limited progress in measuring
accounting quality. At the same time, there is a mounting criticism that the existing models of
discretionary accruals do not properly control for economic performance as well as high quality accruals
(e.g., Wysocki 2009, Dechow et al. 2010, Gerakos 2012). In this paper, I have presented a conceptual
framework and an empirical strategy that enable identification of accounting quality (quality of
earnings/accruals) and also directly addresses the existing criticisms. The strategy is based on our
knowledge of accounting processes, which implies the reversal in both accounting error and high quality
accruals. The information in accruals reversal, in turn, can be used as a source of identifying variation and
naturally calls for the GMM estimation. This source of identification is rather general and does not
depend on a particular set of assumptions about the model.
I also offered a parsimonious empirical model of accruals as a dynamic process. The proposed
characterization is intuitive and reflects institutional features of accrual accounting. Specifically, accruals
are characterized as a sum of two distinct components. This model is based on the idea that both cash
flows and earnings can be viewed as noisy measures of the same underlying economic performance. The
28
first component plays a performance measurement role, in line with Dechow (1994), by resolving the
timing error in cash flows. The second component represents accounting error. Both components exhibit
reversal. Such a characterization makes it natural to characterize accounting quality both in terms of the
magnitude of accounting error introduced via the accrual process (cost of using accruals), and the
magnitude of high quality accruals (benefit of using accrual accounting). The later components have been
largely unaddressed in the literature on accounting quality.
I offer a number of generalizations to the baseline set of assumptions under which the model can
be estimated. Most notably, I offer a new approach to testing for earnings management and discuss how
the model can be modified to explicitly deal with the presence of accounting conservatism. The paper
offers a number of avenues for future research, which should determine which assumptions and model
generalizations are empirically important.
29
References
AICPA. “Objectives of Financial Statements.” Study Group on the Objectives of Fianncial Statements
(1973).
Ball, Ray; and Lakshmanan Shivakumar. “The Role of Accruals in Asymmetrically Timely Gain and
Loss Recognition.” Journal of Accounting Research 44 (2006) 207–42.
Beyer, Anne; Ilan Guttman; and Ivan Marinovic. “Earnings Management and Earnings Quality: Theory
and Evidence.” Stanford University working paper. March 2014.
Dechow, Patricia M. “Accounting Earnings and Cash Flows as Measures of Firm Performance: The Role
of Accounting Accruals.” Journal of Accounting and Economics 18 (1994) 3–42.
Dechow, Patricia M.; and Ilia D. Dichev. “The Quality of Accruals and Earnings: The Role of Accrual
Estimation Errors.” The Accounting Review 77 (2002) 35–59.
Dechow, Patricia; Weili Ge; and Catherine Schrand. “Understanding Earnings Quality: A Review of the
Proxies, their Determinants and their Consequences.” Journal of Accounting and Economics 50
(2010) 344–401.
Dechow, Patricia M.; Amy P. Hutton; Jung Hoon Kim; and Richard G. Sloan. “Detecting Earnings
Management: A New Approach.” Journal of Accounting Research 50 (2012) 275–334.
Dechow, Patricia M.; Richard G. Sloan; and Amy P. Sweeney. “Detecting Earnings Management.” The
Accounting Review 70 (1995) 193–225.
Easton, Peter D.; Trevor S. Harris; and James A. Ohlson. “Aggregate Accounting Earnings Can Explain
Most of Security Returns: The Case of Long Return Intervals.” Journal of Accounting and Economics
15 (1992) 119–42.
Gerakos, Joseph; “Discussion of Detecting Earnings Management: A New Approach.” Journal of
Accounting Research (2012) 1–13.
Gerakos, Joseph, and Andrei Kovrijnykh. “Performance Shocks and Misreporting.” Journal of
Accounting and Economics 56 (2013) 57–72.
Hansen, Lars P. “Large Sample Properties of Generalized Method of Moments Estimators.”
Econometrica 50 (1982) 1029–54.
Hribar, Paul; and D. Craig Nichols. “The Use of Unsigned Earnings Quality Measures in Tests of
Earnings Management.” Journal of Accounting Research 45 (2007) 1017–53.
Jones, Jennifer J. “Earnings Management During Import Relief Investigations.” Journal of Accounting
Research 29 (1991) 193–228.
Nikolaev, Valeri V. Estimating Accounting Quality. Working paper. (2014).
Stickney, Clyde, P.; and L. Weil Roman. “Financial Accounting, an introduction to concepts, methods,
and uses.” 12th edition, Thomson Higher Education, Mason, OH.
Wysocki, Peter D. “Assessing Earnings and Accruals Quality: U.S. and International Evidence.” MIT
working paper. (January 2009).
30